Strata web mining tutorial

1

Scale Unlimited

Web Mining

photo by: i_pinz, ﬂickr
Strata 2012
Copyright (c) 2012 Scale Unlimited. All Rights Reserved. Reproduction or distribution of this document in any form without prior written permission is forbidden.

Tuesday, February 28, 12

2

Welcome to Web Mining!

This class is a tutorial on large scale web mining
Topics covered
Overview of web mining
Web crawling - broad & focused
Text mining - extracting value
Hands-on lab
Tips and traps



3

Meet Your Instructor

Ken Krugler - direct from Nevada City, California
Founder of TransPac Software, Krugle, Bixo Labs/Scale Unlimited
Developer of Bixo web mining toolkit
Committer on Apache Tika
Developer and trainer for Hadoop, Solr and Cascading
Actively web mining for six years



4

Agenda

9:00am - Overview 11:00am - Web Mining Lab
9:30am - Web Crawling 11:45am - Lab Review
10:00am - Text Mining 12:00pm - Summary
10:30am - Break 12:15pm - Q&A



5

Overview

Web Mining

photo by: exfordy, ﬂickr
Strata 2012


6

Key Questions

Which of the three types of web mining are we focusing on today?
What makes web pages “noisy”?



7

What is web mining?



7

What is web mining?

Extracting useful information from the World-wide Web



8

What is web mining?



8

What is web mining?

Web structure - link graph analysis



9

What is web mining?

173.255.195.185 - - [05/Sep/2011:06:03:56 -0600] "GET /feed/ HTTP/1.1" 200 166 "-"
67.124.22.71 - - [05/Sep/2011:06:03:58 -0600] "GET /summary/ HTTP/1.1" 200 809 "-"
89.105.44.90 - - [05/Sep/2011:06:04:02 -0600] "GET /feedx/ HTTP/1.1" 404 0 "-"



9

What is web mining?


173.255.195.185 - - [05/Sep/2011:06:03:56 -0600] "GET /feed/ HTTP/1.1" 200 166 "-"



9

What is web mining?

Web usage - server logs

173.255.195.185 - - [05/Sep/2011:06:03:56 -0600] "GET /feed/ HTTP/1.1" 200 166 "-"



10

What is web mining?



10

What is web mining?

Web content - text and images



11

Web Content Mining

Analyzing data from web pages
Typically three types of page processing
Unstructured - get rid of “boilerplate” text, analyze sentiment
Semi-structured - ﬁnd names of people with phone numbers
Structured - ﬁnd hotel name, address, phone number, reviews
Plus inter-document analysis
Clustering



12

Crawling versus Mining



12


Web mining combines...



12


web crawling - ﬁnding & fetching content



12


data mining - extracting useful information



12



Both ﬁelds are broad and deep - for example



12



optimal crawling strategies



12



machine learning for page classiﬁcation



12



machine learning for page classiﬁcation
automatically extracting structured data



13

What is “Large Scale”?

More than what you can handle with one server
Many single-server solutions for mining web pages
Harder when you include text analytics
And (almost) impossible when you get to 100M+ pages

So you need some kind of distributed processing framework



14

Key Aspects of Web Mining

Crawling - ﬁnding the “good stuff”

Extracting - getting the “right data”

Processing - turning bytes into bucks



15

Finding the “good stuff”

Often feels like “needle in a haystack”
E.g. even 100M pages is 0.1% of total web

Need to optimize time + cost per useful result
Can’t afford to waste time on pages that aren’t useful
And each page has cost to data provider



16

Getting the “right data”

Scale and precision are in opposition
One area of one site can be precision-processed
All areas of 50M domains means you have to be general

Pages are noisy
Ads
Boilerplate (navigation, etc)
SEO



17

Processing the results

1TB data ﬁle has very little value
Actually less value that a small ﬁle that can be opened & viewed
Has to be turned into something with value

Often processing is considered part of web mining
Reduction - turning petabytes into pie charts
Indexing - being able to search the data
Analytics - clustering, training models for recommenders



18

Q&A

Which of the three types of web mining are we focusing on today?
What makes web pages “noisy”?



19

Web Crawling

Web Mining

photo by: Graham Racher, ﬂickr
Strata 2012


20

Key Questions

What are three general types of web crawls?
What can make it hard to accurately score a page?



21

What is “web crawling”?
Includes fetching pages, of course
But also has aspect of spider crawling over a web
Extracting outlinks to discover new pages
Which means parsing the fetched content
Managing state of the crawl
And all of the implicit rules
Robots exclusion protocol
User agent
Request rate



22

Types of web crawls
Broad
Few or no limits to what domains/pages to process
Typically what people think of - Googlebot, bingbot, Baiduspider, ...
Focused
Uses page scoring -> outlinks to guess at quality of unfetched pages
Often has whitelist of domains to avoid traps
Domain
For a limited number of domains
Typically for precise extraction of data



23

The “don’t craw” crawl

Leverage other people’s crawl data
Can be faster, cheaper
Reduces load on servers
Public datasets
Common crawl
Wikipedia - use data dump!
Commercial providers
Spinner, InfoChimps



24

Crawling Solutions

General rule - don’t roll your own!
Easy to make something simple
Hard to make something scalable, robust, efﬁcient
Open source options
Java - Nutch, Heritrix, Bixo, Droids
Python - http://scrapy.org/
PHP - http://astellar.com/php-crawler/



25

What makes it hard?

Web mining breaks the implicit contract with web sites
You often aren’t creating an index that drives trafﬁc to them
So why should they let you use bandwidth & server cycles?
The web is a nearly inﬁnite set of edge cases
Every possible problem will occur, with a broad enough crawl
And not everybody plays nice
Link farms/honeypots, malicious sites, angry webmasters
Plus you have to be able to work at scale



26

Scaling Solutions

Needs to be reliable, scalable, fault tolerant
Single server can fetch lots of pages
But scaling is issue with post-processing
Several options
Hadoop - Nutch, Bixo
Custom queuing system - Heritrix, Droids
Storm - scalable queuing



27

Focused Crawling 101

How to maximize results while minimizing cost
aka Finding Good Stuff Fast
Only crawl pages that you think are likely to be good
Reduces cost through
Less time spent fetching worthless pages
Lower bandwidth/CPU/storage costs
Fewer angry webmasters



28

Focused Crawl Details

Seed URLs - Good starting point
URL State - DB of all known URLs
Page Score - “Quality” of page
Link Score - Page Score/outlinks
Fetched Pages - Saved results



29

Finding Seed URLs

List of all registered domains - Complete, but big (100M+) Broad

DMOZ - lots of spam/porn
Alexa/Quantcast “top sites” list - top 1M US sites by trafﬁc
Wikipedia - use outlink dump if possible
Tweets - with ﬁltering, e.g. Gnip, DataSift
Using search Narrow
Manually entering URLs - slow, but curated
Using API - faster, typically limited, can have junk



30

Scoring Pages

Analyze text on page
Typically means tokenizing text
“The sport of ultimate is..” => “the”, “sport”, “of”, “ultimate”, “is”, ...
Simple term-based
Count occurrences of all phrases, good phrase, bad phrases
Calculate ratios of counts: good/all - bad/all = score



31

Scoring using SVM
SVM = support vector machine
Trained using “documents” that have features, and a class
“good” : “ultimate”, “ultimate frisbee”, “disc”, “sport”, “throw”, “run”
“bad” : “golf”, “timeshare”, “aardvark”, “potato”

Creates a statistical model
Divides all training documents into separate classes
Used to give an unknown document a class



32

Challenges with Scoring
How do you decide that a page is “good”?
Might be mostly graphics with few words
Could be a deﬁnition of the term

Min threshold for amount of real content
Detecting link farms with fake content



33

Chrome, cruft, and boilerplate
Navigational links
Sidebar elements
Ads, SEO links

Can use Boilerpipe & other “cleaners”



34

Expanding the crawl frontier

Have to parse the page to find outlinks
Need to normalize links
http://scaleunlimited.com == http://www.scaleunlimited.com
http://www.scaleunlimited.com == http://www.scaleunlimited.com/
Skipping links to low-value pages
Links to images, pdf files, other binary types (using suffix)
Links to DB-generated pages



35

Focused Domain Crawl

Very speciﬁc, explicit crawl of one domain
Typically involves discovery of target content pages
Often uses URL patterns to synthesize links
Page X in site has list of product; a, b, c, d...
Product pages are <domain>/product/a or b or c or d...



36

Discovery vs Extraction

Focused Domain Crawl has two distinct phases
Crawling to discover details pages
Fetching/processing details pages
Often phases are co-mingled, for efﬁciency
Need to track what kind of page in the URL State DB



37

Goby Crawl Example
http://www.goby.com has information on lots of attractions
http://www.goby.com/boston-ma has list of categories

http://www.goby.com/<category>--near--<city>-<state>
Often need to paginate listing pages, to get all details links



38

Q&A

What are three general types of web crawls?
What can make it hard to accurately score a page?



39

Data Extraction

Web Mining

Strata 2012


40

Key Questions

What are the three general approaches for data extraction?
Why might you want to detect the language of a page?



41

You’ve got a page, now what?

Time to extract the data you need
Three attributes of extraction, pick any two
Broad - across lots of domains and page formats
Precise - very speciﬁc types of data
Accurate - low error rate
Three general approaches
Unstructured (broad, accurate) - “just text”
Semi-structured (broad, precise) - ﬁnding meaning in text
Structured (precise, accurate) - getting exactly the data you need



42

Common Tasks - Cleaning

The HTML needs to be cleaned up
Lots of messy data, especially when hand-edited
Even HTML (2.0? 3.2? 4.0.1?) should be converted to XHTML
Various libraries help with “cleaning” the HTML
TagSoup, NekoHTML, HtmlCleaner
Note that end result won’t match original text



43

Common Tasks - Charset

You get bytes back from the web server
You need a charset to convert bytes to characters
HTTP response header - “Content-Type: text/html; charset=UTF-8”
HTML meta tag - <meta http-equiv=”Content-Type” content=“...” />
Analysis of text - byte sequence statistics
Several packages support this
Tika, ICU



44

Common Tasks - Link Extraction

Needed to have a crawl - where new links come from
Means you need XHTML so you can parse the markup
Not just <a href=“xxx”>
img, frame, iframe, link, map, area



45

Common Tasks - Boilerplate

For unstructured and semi-structured
Can improve the quality of results
Especially important for machine learning
Boilerplate text can dramatically skew statistics
Creates a noisier signal



46

Common Tasks - Language
Often used for ﬁltering or alternative processing
Target audience is only interested in Spanish
I need to tokenize Japanese differently
Clustering improves when it’s segmented by language
Multiple signals for selecting language, same as charset detection
HTTP response header: Content-Language: es
HTML meta tag - <meta http-equiv=”Content-Language” content=“es” />
HTML tag attributes - <html lang=“es”>
Analysis of text - ngram statistics, short words



47

Unstructured Extraction

Goal is extracting text, without much additional processing
Often has a few ﬁelds, from HTML
Title - from <head><title>The title of my page</title></head>
Description - from <meta name=“description” content=“ultimate frisbee” />
Body - from <body>...all elements that contain text, like <p>...</body>



48

Semi-structured Extraction

Goal is ﬁnding structured data in random text
Can be applied broadly, since it’s not (very) format-speciﬁc
Accuracy suffers, because of breadth of input data formats
Beware the academic algorithm
Examples of what does work...
Easy patterns: telephone numbers, dates
Microformats: hCalendar, hCard, hReview, ...
Natural Language Processing (NLP): named entities



49

Structured Extraction

Precise extraction of speciﬁc types of data
Typically is to one area of one site
Often handled with XPath, and maybe regular expressions
//div[@id=‘<id of target div>’]/p
div and span are beautiful tags
Commonly used with CSS
Which means they are (more) stable



50

How to ﬁgure out XPath

Firebug is your friend
Plug-in for Firefox
Will show you the full XPath for each element
Note that browsers will re-write HTML (e.g. tbody element)
The DOM you see is often generated with Javascript



51

XPath demo

Firebug
XPath tool



52

Dealing with Javascript

Required if page generates target content using JS
Forces you to use Firebug or equivalent to inspect the DOM
Options for processing include...
HtmlUnit
qt-webkit
headless Mozilla



53

Javascript challenges

10x slower than just loading the page text
Good way to make a webmaster angry
Lots of extra load on server
Can skew website stats
Often has issues
Pages that work in FF or IE but not HtmlUnit
Pages that cause HtmlUnit to hang



54

Q&A

What are the three general approaches for data extraction?
Why might you want to detect the language of a page?



55

Web Crawling Lab

Web Mining

Strata 2012


56

Key Questions

Were you able to build and run the code locally?
Were you able to run a crawl in Elastic MapReduce?
Were you able to improve the focused crawl?



57

ImageFinder Details

Find images about Ultimate Frisbee
Focused crawl
Fixed list of seed URLs
Positive & negative terms used to score pages
Extract images from page



58

Running ImageFinder

Can run locally, with restrictions
Only one fetcher thread
Only 5 pages/loop
Can run in Hadoop cluster
Amazon Elastic MapReduce
CrawlRunner uploads job jar, creates “Job Flow Step”
Limited to 100 pages/loop, 2 loops
Will take up to 10 minutes to run loops



59

Running in Elastic MapReduce

Watch your job via http://strata.scaleunlimited.com:9100/
Your job name will include your username
Results get added to searchable index
http://strata.scaleunlimited.com/solr/strata/
Search for student:<username> to ﬁnd your results



60

Solr Results Issue
Note links vs. images
These are “image” URLs that are actually to pages
Double-bonus on exercise
...ﬁx this problem :)



61

Lab Details

Code is in strata-web-mining folder you downloaded
Details of code in strata-web-mining/doc/README-Description
Instructions are in strata-web-mining/doc/README-Lab
Please follow the lab steps carefully
Missing a step will cause pain and suffering later



62

Lab Exercises

First goal is to build code and run locally
Next is to build code and run in real cluster
Then you get to try to optimize the focused crawl
And (if you’re fast) try ﬁnding images for a different topic



63

Trouble-shooting & Timing

I’ll be walking around - raise your hand if you need help
But with 100+ people, I’ll be talking fast :)
We’ve got an hour (or more) before summary/Q&A
Have fun...



64

Q&A

Were you able to build and run the code locally?
Were you able to run a crawl in Elastic MapReduce?
Were you able to improve the focused crawl?



65

Summary & QA

Web Mining

Strata 2012


66

Key Challenges

Complexity of large scale web crawling

Challenges with extracting the right data

Extra work to turn results into value



67

Ethical Crawling

Always have a real, valid, informative user agent name
Always honor the robot exclusion protocol - robots.txt
Limit your crawl rate - parallelism, crawl delay, pages/day
Immediately comply with blacklisting and data removal requests



68

Avoiding getting blocked

Follow all ethical crawling guidelines
Gradually ramp up your crawl rate
Gives webmasters time to complain before it’s a serious problem
Avoid Javascript if at all possible
Don’t follow form links
Grovel shamelessly



69

Resources

Hadoop - http://hadoop.apache.org
Cascading - http://www.cascading.org
Bixo - http://openbixo.org
Web Data Mining by Bing Liu



70

Question?

I might have answers

ken@scaleunlimited.com
@kkrugler



Strata web mining tutorial

Recommended

Recommended

More Related Content

Viewers also liked

Viewers also liked (17)

Similar to Strata web mining tutorial

Similar to Strata web mining tutorial (19)

More from Ken Krugler

More from Ken Krugler (8)

Recently uploaded

Recently uploaded (20)

Strata web mining tutorial